[FFmpeg-devel] [PATCH v3] avformat/mov: add option max_stts_delta

"zhilizhao(赵志立)" quinkblack at foxmail.com
Sun Dec 26 19:39:16 EET 2021



> On Dec 27, 2021, at 12:00 AM, Gyan Doshi <ffmpeg at gyani.pro> wrote:
> 
> Very high stts sample deltas may occasionally be intended but usually
> they are written in error or used to store a negative value for dts correction
> when treated as signed 32-bit integers.
> 
> This option lets the user set an upper limit, beyond which the delta is clamped to 1.
> Values greater than the limit if negative when cast to int32 are used to adjust onward dts.
> 
> Unit is the track time scale. Default is UINT_MAX - 48000*10 which
> allows upto a 10 second dts correction for 48 kHz audio streams while
> accommodating 99.9% of uint32 range.
> ---
> v3 changes:
> 
> factored out loop
> simplified correction logic
> added doc
> 
> doc/demuxers.texi  |  9 ++++++++
> libavformat/isom.h |  1 +
> libavformat/mov.c  | 55 +++++++++++++++++++++++++---------------------
> 3 files changed, 40 insertions(+), 25 deletions(-)
> 
> diff --git a/doc/demuxers.texi b/doc/demuxers.texi
> index cab8a7072c..99d4ab7bc5 100644
> --- a/doc/demuxers.texi
> +++ b/doc/demuxers.texi
> @@ -713,6 +713,15 @@ specify.
> 
> @item decryption_key
> 16-byte key, in hex, to decrypt files encrypted using ISO Common Encryption (CENC/AES-128 CTR; ISO/IEC 23001-7).
> +
> + at item max_stts_delta
> +Very high sample deltas written in a trak's stts box may occasionally be intended but usually they are written in
> +error or used to store a negative value for dts correction when treated as signed 32-bit integers. This option lets
> +the user set an upper limit, beyond which the delta is clamped to 1. Values greater than the limit if negative when
> +cast to int32 are used to adjust onward dts.
> +
> +Unit is the track time scale. Range is 0 to UINT_MAX. Default is @code{UINT_MAX - 48000*10} which allows upto
> +a 10 second dts correction for 48 kHz audio streams while accommodating 99.9% of @code{uint32} range.
> @end table
> 
> @subsection Audible AAX
> diff --git a/libavformat/isom.h b/libavformat/isom.h
> index ef8f19b18c..625dea8421 100644
> --- a/libavformat/isom.h
> +++ b/libavformat/isom.h
> @@ -305,6 +305,7 @@ typedef struct MOVContext {
>     int32_t movie_display_matrix[3][3]; ///< display matrix from mvhd
>     int have_read_mfra_size;
>     uint32_t mfra_size;
> +    uint32_t max_stts_delta;
> } MOVContext;
> 
> int ff_mp4_read_descr_len(AVIOContext *pb);
> diff --git a/libavformat/mov.c b/libavformat/mov.c
> index 2aed6e80ef..f93c8bef31 100644
> --- a/libavformat/mov.c
> +++ b/libavformat/mov.c
> @@ -2925,6 +2925,8 @@ static int mov_read_stts(MOVContext *c, AVIOContext *pb, MOVAtom atom)
>     unsigned int i, entries, alloc_size = 0;
>     int64_t duration = 0;
>     int64_t total_sample_count = 0;
> +    int64_t current_dts = 0;
> +    int64_t corrected_dts = 0;
> 
>     if (c->fc->nb_streams < 1)
>         return 0;
> @@ -2965,11 +2967,34 @@ static int mov_read_stts(MOVContext *c, AVIOContext *pb, MOVAtom atom)
>         sc->stts_data[i].count= sample_count;
>         sc->stts_data[i].duration= sample_duration;
> 
> -        av_log(c->fc, AV_LOG_TRACE, "sample_count=%d, sample_duration=%d\n",
> +        av_log(c->fc, AV_LOG_TRACE, "sample_count=%u, sample_duration=%u\n",
>                 sample_count, sample_duration);
> 
> -        duration+=(int64_t)sample_duration*(uint64_t)sample_count;
> -        total_sample_count+=sample_count;
> +        /* STTS sample offsets are uint32 but some files store it as int32
> +         * with negative values used to correct DTS delays.
> +           There may be abnormally large values as well. */
> +        if (sample_duration > c->max_stts_delta) {
> +            // assume high delta is a correction if negative when cast as int32
> +            int32_t delta_magnitude = (int32_t)sample_duration;
> +            av_log(c->fc, AV_LOG_WARNING, "Too large sample offset %u in stts entry %u with count %u in st:%d. Clipping to 1.\n",
> +                   sample_duration, i, sample_count, st->index);
> +            sc->stts_data[i].duration = 1;
> +            corrected_dts += (delta_magnitude < 0 ? (int64_t)delta_magnitude : 1) * sample_count;
> +        } else {
> +            corrected_dts += sample_duration * sample_count;
> +        }
> +
> +        current_dts += sc->stts_data[i].duration * sample_count;
> +
> +        if (current_dts > corrected_dts) {
> +            int64_t drift = (current_dts - corrected_dts)/sample_count;
> +            uint32_t correction = (sc->stts_data[i].duration > drift) ? drift : sc->stts_data[i].duration - 1;
> +            current_dts -= correction * sample_count;
> +            sc->stts_data[i].duration -= correction;
> +        }
> +
> +        duration+=(int64_t)sc->stts_data[i].duration*(uint64_t)sc->stts_data[i].count;

The second cast doesn’t help much.

> +        total_sample_count+=sc->stts_data[i].count;
>     }
> 
>     sc->stts_count = i;
> @@ -3856,13 +3881,10 @@ static void mov_build_index(MOVContext *mov, AVStream *st)
>         unsigned int distance = 0;
>         unsigned int rap_group_index = 0;
>         unsigned int rap_group_sample = 0;
> -        int64_t last_dts = 0;
> -        int64_t dts_correction = 0;
>         int rap_group_present = sc->rap_group_count && sc->rap_group;
>         int key_off = (sc->keyframe_count && sc->keyframes[0] > 0) || (sc->stps_count && sc->stps_data[0] > 0);
> 
>         current_dts -= sc->dts_shift;
> -        last_dts     = current_dts;
> 
>         if (!sc->sample_count || sti->nb_index_entries)
>             return;
> @@ -3973,26 +3995,8 @@ static void mov_build_index(MOVContext *mov, AVStream *st)
>                 current_offset += sample_size;
>                 stream_size += sample_size;
> 
> -                /* A negative sample duration is invalid based on the spec,
> -                 * but some samples need it to correct the DTS. */
> -                if (sc->stts_data[stts_index].duration < 0) {
> -                    av_log(mov->fc, AV_LOG_WARNING,
> -                           "Invalid SampleDelta %d in STTS, at %d st:%d\n",
> -                           sc->stts_data[stts_index].duration, stts_index,
> -                           st->index);
> -                    dts_correction += sc->stts_data[stts_index].duration - 1;
> -                    sc->stts_data[stts_index].duration = 1;
> -                }
>                 current_dts += sc->stts_data[stts_index].duration;
> -                if (!dts_correction || current_dts + dts_correction > last_dts) {
> -                    current_dts += dts_correction;
> -                    dts_correction = 0;
> -                } else {
> -                    /* Avoid creating non-monotonous DTS */
> -                    dts_correction += current_dts - last_dts - 1;
> -                    current_dts = last_dts + 1;
> -                }
> -                last_dts = current_dts;
> +
>                 distance++;
>                 stts_sample++;
>                 current_sample++;
> @@ -8577,6 +8581,7 @@ static const AVOption mov_options[] = {
>     { "decryption_key", "The media decryption key (hex)", OFFSET(decryption_key), AV_OPT_TYPE_BINARY, .flags = AV_OPT_FLAG_DECODING_PARAM },
>     { "enable_drefs", "Enable external track support.", OFFSET(enable_drefs), AV_OPT_TYPE_BOOL,
>         {.i64 = 0}, 0, 1, FLAGS },
> +    { "max_stts_delta", "treat offsets above this value as invalid", OFFSET(max_stts_delta), AV_OPT_TYPE_INT, {.i64 = UINT_MAX-48000*10 }, 0, UINT_MAX, .flags = AV_OPT_FLAG_DECODING_PARAM },

It works, but can be a little confusing for int type to has a default
value > INT_MAX and upper limit UINT_MAX. There are other such usage
in the code base, maybe it’s OK.

> 
>     { NULL },
> };
> -- 
> 2.33.0
> 
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".



More information about the ffmpeg-devel mailing list